Visual inspection aid

Abstract

A visual inspection aid is provided that can further enhance the reliability of visual inspection. The visual inspection aid comprises a conveyor that conveys a tablet to be inspected visually, a camera (83, 84) that captures the conveyed tablet, an inspection unit (858) that determines the quality of the tablet based on image data obtained by the camera (83, 84), a display unit (85b, 85c) positioned at a location visible to an inspector performing a visual inspection, a display control unit (854) capable of displaying information to identify defective tablets on the display unit (85b, 85c), and a conveyor control unit (7) that temporarily stops the conveying of the tablet by the conveyor (4) when a defect is determined by the inspection unit (858). The inspector can effectively use the display content of the display unit (85b, 85c) to easily identify defective tablets and can also more accurately remove defective tablets that are temporarily stopped. For this reason, the reliability of visual inspection can be increased.

Description

Technology Field

[0001] The present invention relates to a visual inspection aid used for visual inspection of tablets. Background Technology

[0002] Conventionally, it is widely practiced to visually inspect objects being returned for inspection. As a device used to visually inspect a tablet as an object for inspection, it has been proposed to have an inversion section that inverts the front and back sides of the tablet in the middle of the tablet's return path (see, for example, Patent Document 1 et al.). In this device, it is possible to visually inspect one side of the tablet's front or back side upstream of the inversion section, and the other side of the tablet's front or back side downstream of the inversion section, respectively, so that one inspector can inspect both the front and back sides of the tablet. Prior art literature

[0003] Japanese Patent Publication No. 2001-95898 The problem to be solved

[0004] However, in the above-mentioned device, since tablet inspection is performed only by visual inspection, there is a risk of overlooking defective tablets. Furthermore, although an inspector who discovers a defective tablet removes it from the return path, there is a risk of damaging the defective tablet or mistakenly removing a good tablet that is different from the defective one. If such oversight, error, or mistake occurs, the defective tablet is discharged downstream.

[0005] The present invention has been made in consideration of the above circumstances, and its purpose is to provide a visual inspection aid that can more reliably prevent the occurrence of such oversight, errors, or mistakes, and can further enhance the reliability of visual inspection. means of solving the problem

[0006] Below, each means suitable for solving the above-mentioned objective is described in separate paragraphs. Additionally, specific effects specific to the corresponding means are noted as necessary.

[0007] Means 1. A visual inspection auxiliary device equipped with a conveying means for conveying tablets subject to visual inspection, wherein

[0008] An imaging means for imaging a tablet being transported by the above-mentioned transport means, and

[0009] A determination means for determining the quality of a tablet based on image data obtained by the above-mentioned imaging means, and

[0010] A display means placed in a position visible to an inspector performing a visual inspection of the tablet, and

[0011] In the above-mentioned display means, a display control means capable of displaying information for specifying a tablet that has been judged defective by at least the above-mentioned judgment means, and

[0012] A visual inspection assist device characterized by having a return control means for temporarily stopping the return of tablets by the return means when a defect is determined by the above determination means, and configured to be able to identify and remove the tablets determined to be defective by using the display content in the above display means.

[0013] According to the above means 1, the quality of the tablet being returned by the return means (i.e., subject to visual inspection) is determined by the judgment means. And, if a defect is determined by the judgment means, the return of the tablet is temporarily stopped. In addition, in the display means, information for identifying the tablet that has been determined to be defective by the judgment means is displayed.

[0014] Therefore, an inspector performing a visual inspection can effectively utilize the marking content of the marking means to easily identify a defective tablet (defective tablet) among a plurality of tablets. In addition, since the tablet is temporarily stopped when a defect is determined, the inspector can remove the identified defective tablet more accurately. This makes it more effective to prevent overlooking a defective tablet, misidentifying a defective tablet, or mistakenly removing a good tablet that is different from the defective one. As a result, it makes it more effective to prevent defective tablets from flowing downstream and increases the reliability of the visual inspection. Mean 2.A visual inspection assisting device equipped with a conveying means for conveying a tablet subject to visual inspection, the device comprises: an imaging means for capturing an image of a tablet conveyed by the conveying means; a judgment means for determining the quality of a tablet based on image data obtained by the imaging means; a display means positioned at a location visible to an inspector performing a visual inspection of the tablet; a display control means capable of displaying information for identifying a tablet that has been judged defective by at least the judgment means in the display means; and a conveying control means for temporarily stopping the conveying of the tablet by the conveying means when a defect is judged by the judgment means; and an imaging control means for controlling the imaging means, wherein the imaging control means controls the imaging means to capture an area corresponding to at least the tablet judged defective during the temporary suspension of the tablet to obtain image data for re-evaluation, and simultaneously, the judgment means determines the quality of the tablet based on the image data for re-evaluation, and the conveying control means causes the conveying of the tablet that is temporarily suspended to resume. A visual inspection assistance device characterized by the ability to restart the above-mentioned return means, and the restart of the above-mentioned return means by the above-mentioned return control means being configured to be carried out on the condition that the judgment result by the above-mentioned judgment means based on the above-mentioned re-judgment image data is good. According to the above-mentioned means 2, when a defective tablet is additionally discovered and the return of the tablet is temporarily stopped, it is possible to restart the return means and resume the return of the tablet only when the judgment result based on the re-judgment image data obtained by capturing at least the range corresponding to the defective tablet (i.e., the range including the part where the defective tablet is located) is good. Therefore, even if a defective tablet is overlooked or a mistake occurs, it is possible to more reliably prevent the defective tablet from being accidentally discharged downstream.As a result, the reliability of the visual inspection can be further improved. Mean 3. A visual inspection auxiliary device equipped with a conveying means for conveying a tablet subject to visual inspection, the device comprises: an imaging means for capturing an image of a tablet conveyed by the conveying means; a judgment means for determining the quality of a tablet based on image data obtained by the imaging means; a display means positioned at a location visible to an inspector performing a visual inspection of the tablet; a display control means capable of displaying information for identifying a tablet that has been judged defective by at least the judgment means in the display means; and a conveying control means for temporarily stopping the conveying of the tablet by the conveying means when a defect is judged by the judgment means. The conveying means conveys the tablets in a state arranged in multiple rows. The judgment means comprises an identification means generated by training only learning data based on image data of a good tablet for a neural network having an encoding unit that extracts feature quantities from input image data and a decoding unit that reconstructs image data from the feature quantities, and the A visual inspection assist device characterized by comprising: a means for acquiring reconstructed image data capable of acquiring reconstructed image data as reconstructed image data by inputting original image data based on image data obtained by an imaging means into the identification means; a comparison means capable of comparing the original image data and the reconstructed image data; and configured to determine the quality of a tablet based on a comparison result by the comparison means, wherein the learning data is image data obtained by extracting an area corresponding to a plurality of tablets in one column from image data corresponding to a good tablet for each column of tablets, and the original image data is image data obtained by extracting an area corresponding to a plurality of tablets in one column from image data obtained by the imaging means for each column of tablets.In addition, "image data resulting from the refinement of good products" may include image data resulting from the refinement of good products accumulated from previous quality judgments, or virtual good product image data generated using the above image data. Furthermore, the above "neural network" includes, for example, a convolutional neural network having multiple convolutional layers. The above "learning" includes, for example, deep learning. The above "identification means (generation model)" includes, for example, an autoencoder or a convolutional autoencoder. Furthermore, since the "identification means" is generated by learning only the image data resulting from the refinement of good products, the reconstructed image data generated when the original image data resulting from defect refinement is input into the identification means will be almost identical to the original image data from which noise (defective parts) has been removed. That is, when there are defective parts in the refinement, virtual image data based on the refinement is generated as reconstructed image data, assuming there are no defective parts. However, as training data used for learning identification means or original image data used for comparison (i.e., pass / fail determination) by comparison means, one can consider using, for example, image data corresponding to the entire field of view of the imaging means (global data) or image data obtained by extracting the area occupied by each refinement one by one from the global data (individual refinement data). However, if global data is used, the amount of data required for training becomes massive, which may lead to increased costs associated with training. Furthermore, if individual refinement data is to be used, since the shape or size of each refinement differs slightly, very cumbersome settings must be performed to extract individual refinement data, or the extraction process may require a considerable amount of time, which may consequently make extraction processing difficult.In this regard, according to the above means 3, additionally, as training data or original image data, image data obtained by extracting areas corresponding to multiple tablets in one column (adjacent to each other in the return direction) for each column of tablets is used. Therefore, the amount of data required for training can be reduced relatively, thereby reducing the cost associated with training. In addition, the extraction processing of training data or original image data can be made easier. Furthermore, according to the above means 3, the original image data and the reconstructed image data, which is reconstructed by inputting the original image data into the identification means, are compared, and the quality of the tablet is determined based on the comparison result. Therefore, the two image data being compared each relate to the same tablet. Consequently, since the shape or appearance of the tablet in the two image data being compared becomes almost identical, there is no need to set relatively loose judgment conditions to prevent false detection due to differences in shape or appearance, and stricter judgment conditions can be set. In addition, the imaging conditions (e.g., the placement position or angle of the tablets, the brightness and contrast state, or the camera's field of view) can be matched in the two image data being compared. As a result, the quality judgment of the tablets can be performed very accurately.

[0015] Mean 4. A visual inspection aid device described in any one of means 1 to 3, characterized in that the display control means is configured to be capable of displaying information for identifying a tablet that has been judged defective by at least the judgment means, in correspondence with the image data corresponding to the tablet.

[0016] According to the above means 4, in the display means, information for identifying a tablet that has been determined to be defective by the determination means is displayed in correspondence with image data corresponding to the tablet (for example, a specific mark indicating that it is a defective product is displayed in a state overlapping with the defective tablet in the image data, etc.). Accordingly, identifying the defective tablet becomes easier, and the effort or time required to remove the defective tablet can be further reduced.

[0017] delete

[0018] delete

[0019] delete

[0020] delete

[0021] delete

[0022] delete

[0023] Means 5. The tablet has a surface and a back side, and

[0024] The above-mentioned conveying means has a transparent or translucent permeable portion installed in at least a part thereof, and at the same time has a tablet support portion that supports the tablet being conveyed, and

[0025] The above imaging means comprises a first imaging means and a second imaging means positioned at a location where the tablet support is fitted upper and lower, and

[0026] One side of the surface and the back side of the tablet is imaged by the first imaging means, and the other side of the surface and the back side of the tablet is imaged through the transmission part by the second imaging means.

[0027] A visual inspection aid device described in any one of means 1 to 3, characterized in that the above determination means is configured to determine the quality of a tablet based on image data according to the surface and back surface of the tablet obtained by the above first imaging means and the above second imaging means.

[0028] According to the above means 5, the quality of the surface and back side of the tablet can be determined using a permeable part, so the quality of the tablet can be determined more precisely. In addition, since the second imaging means is positioned below the tablet support part, it is possible to prevent the second imaging means from interfering with visual inspection.

[0029] In addition, when the above means 5 is applied to the above means 2, the operation of the return means may be permitted only when both the surface and the back side of the tablet are determined to be good based on image data for re-evaluation. Therefore, even if a defective tablet is mishandled and happens to be overturned, the operation of the return means is not permitted. Thus, the outflow of defective tablets downstream can be very effectively prevented.

[0030] delete

[0031] delete

[0032] delete

[0033] delete

[0034] delete

[0035] delete

[0036] delete

[0037] delete

[0038] delete

[0039] delete

[0040] delete

[0041] delete

[0042] delete

[0043] delete

[0044] Means 6. The tablet has a surface and a back surface, and a side surface located between the surface and the back surface, and

[0045] The above-mentioned return means has a reversing part that reverses the front and back sides of the returned tablet, and

[0046] The above imaging means comprises a side imaging means for imaging the side of a tablet located in the inversion section, and

[0047] A visual inspection aid device described in any one of means 1 to 3, characterized in that the above determination means is configured to determine the quality of the side of the tablet based on image data obtained by the above side imaging means.

[0048] According to the above means 6, since an inversion part that inverts the front and back sides of the tablet is installed, visual inspection of the front and back sides of the tablet can be easily performed.

[0049] In addition, by utilizing the inversion unit, a quality determination can be performed regarding the side of the tablet. Therefore, the precision of the quality determination can be further increased. Furthermore, the outflow of defective tablets downstream can be prevented more reliably, thereby increasing the reliability of visual inspection.

[0050] In addition, technical details according to each of the aforementioned means can be appropriately combined. Brief explanation of the drawing

[0051] Figure 1 is a schematic diagram showing the general configuration of a visual inspection aid device, etc. Figure 2 is a perspective schematic diagram showing a conveying device, etc. Figure 3 is a side view of the tablet. Figure 4 is a block diagram showing the functional configuration of the judgment unit. Figure 5 is a planar schematic diagram showing the returned tablet and test area, etc. Figure 6 is a schematic diagram showing image data obtained by a camera. Figure 7 is a schematic diagram showing original image data. Figure 8 is a schematic diagram representing the training data. Figure 9 is a schematic diagram illustrating the structure of a neural network. Figure 10 is a flowchart showing the flow of the learning process of a neural network. Figure 11 is a flowchart showing the flow of processing when determining the quality of a tablet. Figure 12 is a flowchart showing the flow of the pass / fail judgment process. FIG. 13 is a schematic diagram showing an example of information displayed in the display unit when the tablet is determined to be defective. FIG. 14 is a schematic diagram showing the general configuration of a visual inspection aid device, etc., equipped with a transparent part in another embodiment. FIG. 15 is a perspective schematic diagram showing a conveying device, etc., with a transparent part installed in another embodiment. FIG. 16 is a schematic perspective view showing a side camera, etc., for capturing the side of a tablet in another embodiment. FIG. 17 is a schematic diagram showing an example of information displayed in a display unit in another embodiment. FIG. 18 is a schematic diagram showing an example of information displayed in a display unit in another embodiment. FIG. 19 is a schematic diagram showing an example of information displayed in a display unit in another embodiment. Specific details for implementing the invention

[0052] Hereinafter, an embodiment will be described with reference to the drawings. A visual inspection aid is a device used to perform a visual inspection of a tablet (9) (see FIG. 3) being transported. Additionally, the tablet (9) is a flat plate having a surface (9a) and a back surface (9b), and a side surface (9c) located between these two surfaces (9a, 9b) (see FIG. 3).

[0053] As illustrated in FIG. 1, the visual inspection aid (1) is equipped with a tablet supply device (2), a supply tram (3), a conveying device (4), and a chute (5). In this embodiment, the conveying device (4) constitutes the "conveying means." Additionally, a counter (6a) and a bottle filling device (6b) are installed downstream of the chute (5), and the tablets (9) that have undergone visual inspection are counted by the counter (6a), and the tablets (9) are filled into containers by the bottle filling device (6b).

[0054] The tablet supply device (2) is, for example, a hopper and intermittently supplies tablets (9) to the supply tram (3).

[0055] The supply tram (3) aligns and transports the tablets (9) toward the transport device (4). The supply tram (3) is equipped with a predetermined first vibrator (3a), and the tablets (9) on the supply tram (3) are sent to the transport device (4) by the operation of the first vibrator (3a).

[0056] The conveying device (4) conveys tablets (9) subject to visual inspection in a state where they are arranged in multiple rows. As shown in FIG. 2, the conveying device (4) is equipped with an upstream conveying section (41), an inversion section (42), a downstream conveying section (43), and a second vibrator (44). Additionally, inspectors performing visual inspection are positioned, for example, one person each on the sides of the upstream conveying section (41) and the downstream conveying section (43). In addition, depending on the case, inspectors may be positioned on both sides of the upstream conveying section (41) or the downstream conveying section (43) by fitting them together. That is, a total of four or more inspectors are positioned. Also, FIG. 2 and others schematically show only some of the multiple tablets (9) being conveyed.

[0057] The upstream conveyor (41) continuously conveys the tablet (9) to the downstream side in a state where one of the surface (9a) and the back side (9b) becomes the upper surface. The upstream conveyor (41) has a plurality of upstream conveyor grooves (41a) installed in parallel to send the tablet (9) to the downstream side in an aligned state. When conveying by the upstream conveyor (41), one of the surface (9a) and the back side (9b) of the tablet (9) is visually inspected.

[0058] The inversion section (42) is intended to invert the front and back sides of the tablet (9) while continuously conveying the tablet (9). The inversion section (42) is provided with a plurality of intermediate conveying grooves (42a) that are continuous with the upstream conveying groove (41a), and these intermediate conveying grooves (42a) have a shape that gradually twists toward the downstream side. Accordingly, the tablet (9) conveyed in the upstream conveying section (41) with one of the surface (9a) and the back side (9b) becoming the upper surface passes through the intermediate conveying groove (42a) and is sent to the downstream conveying section (43) with the other of the surface (9a) and the back side (9b) becoming the upper surface.

[0059] The downstream conveyor (43) continuously conveys the tablet (9) to the downstream side with the surface (9a) and the back side (9b) of the upstream conveyor (41) being the back side. In the downstream conveyor (43), a plurality of downstream conveyor grooves (43a) are installed in parallel to the intermediate conveyor groove (42a) to send the tablet (9) to the chute (5) in an aligned state. When conveying by the downstream conveyor (43), the side opposite to the side of the surface (9a) and the back side (9b) of the tablet (9) that was visually inspected in the upstream conveyor (41) is visually inspected. The tablet (9) that has passed through the downstream conveyor (43) flows down the chute (5) and is sent to the counter (6a).

[0060] In addition, in this embodiment, the tablet support part (45), which forms the bottom of the return groove (41a, 42a, 43a) and supports the lower surface of the tablet (9) from below, is configured to have an upper surface with an inclined shape that gradually descends toward the downstream side.

[0061] The second vibrator (44) is a device for applying vibration to the upstream conveyor section (41), the inversion section (42), and the downstream conveyor section (43). As vibration is applied from the second vibrator (44) to the upstream conveyor section (41), the tablets (9) located in each groove section (41a, 42a, 43a) are conveyed to the downstream side. Meanwhile, by stopping the operation of the second vibrator (44), the conveying of the tablets (9) is stopped.

[0062] In addition, the visual inspection aid (1) is equipped with a return control device (7) and a judgment unit (8) as shown in FIG. 1 and 4, in addition to the return device (4), etc. In this embodiment, the return control device (7) constitutes a "return control means."

[0063] First, the return control device (7) will be described. The return control device (7) has the function of initiating (resuming) or temporarily stopping the return of the tablet (9) in the return device (4) by controlling the operation of the return device (4) (particularly the second vibrator (44)).

[0064] The return control device (7) is configured to communicate with the control system (85) described later of the judgment unit (8), and is capable of determining the quality of the tablet (9) by the control system (85) (particularly the inspection unit (858) described later). Furthermore, when the control system (85) determines that the tablet (9) is defective, the return control device (7) temporarily stops the return of the tablet (9) by the return device (4). To describe above, in this embodiment, one tablet (9) is captured at least twice by each camera (83, 84) described later. When the return control device (7) determines that the tablet (9) is defective based on the image data obtained from the first capture of the tablet (9) by the camera (83, 84), the return device (4) controls the return device (4) so ​​that the return of the tablet (9) is temporarily stopped at the timing of the second capture of the tablet (9).

[0065] Additionally, a predetermined return operation unit (not shown) is connected to the return control device (7), and, for example, an inspector performing a visual inspection operates the return operation unit to switch between starting (restarting) and temporarily stopping the return device (4), thereby making it possible to start (restart) or temporarily stop the return of the tablet (9). However, when the return of the tablet (9) is temporarily stopped due to a determination of defect in the tablet (9) as described above, the return control device (7) is configured to allow the return of the return device (4) by operating the return operation unit only when the condition is satisfied that the judgment result is "good" based on image data (image data for re-determination described later) obtained by re-capturing the area corresponding to the defective tablet (9) [hereinafter referred to simply as "defective tablet (9x)" (see FIG. 13)]. Therefore, when the return of the tablet (9) is temporarily suspended due to a defect judgment of the tablet (9), the return of the tablet (9) cannot be resumed by operating the return operation unit alone.

[0066] Next, the judgment unit (8) is described. The judgment unit (8) is equipped with a first lighting device (81), a second lighting device (82), a first camera (83), a second camera (84), and a control system (85). In this embodiment, the first camera (83) constitutes the "first imaging means," and the second camera (84) constitutes the "second imaging means." Additionally, the first camera (83) and the second camera (84) constitute the "imaging means."

[0067] The first lighting device (81) and the second lighting device (82) irradiate a predetermined light (e.g., white light, etc.) from above onto a predetermined inspection area (KA) (see FIG. 5) that includes at least all of the conveyed tablets (9). The first lighting device (81) irradiates light onto the inspection area (KA) in the upstream conveying section (41), and the second lighting device (82) irradiates light onto the inspection area (KA) in the downstream conveying section (43).

[0068] The first camera (83) and the second camera (84) capture an image of the inspection area (KA) from almost directly above. Each camera (83, 84) has an imaging element, such as a CCD (Charge Coupled Device) type image sensor or a CMOS (Complementary Metal Oxide Semiconductor) type image sensor, and an optical system (such as a lens unit or aperture) that forms an image of the object to be captured with respect to the imaging element. Of course, imaging elements other than those described above may be employed. Additionally, each camera (83, 84) is positioned so that its optical axis is orthogonal to the upper surface of the tablet support (45).

[0069] Each camera (83, 84) is driven and controlled by a control system (85) (particularly the camera control unit (853) described later). By capturing the tablet (9) by the camera (83, 84), global data (AD) (see FIG. 6), which is image data representing the entire field of view of the camera (83, 84), is acquired. In this embodiment, the entire field of view of the camera (83, 84) corresponds to the inspection area (KA).

[0070] Image data generated by each camera (83, 84) is converted into a digital signal inside each camera (83, 84) and then transmitted in the form of a digital signal to a control system (85) (particularly an image acquisition unit (855) described later) and stored. Then, the control system (85) performs various image processing or computation processing described later based on the image data.

[0071] The control system (85) is composed of a computer, input / output device and display device, etc., including a CPU (Central Processing Unit) that performs a predetermined operation, a ROM (Read Only Memory) that stores various programs or fixed value data, a RAM (Random Access Memory) that temporarily stores various data when various operation processes are performed, and peripheral circuits thereof.

[0072] And, the control system (85) functions as various functional units such as the main control unit (851), lighting control unit (852), camera control unit (853), display control unit (854), image acquisition unit (855), data processing unit (856), learning unit (857), and inspection unit (858), as the CPU operates according to various programs.

[0073] However, the above various functional parts are realized by the cooperation of various hardware such as the CPU, ROM, and RAM, and there is no need to clearly distinguish between functions realized in hardware or software, and some or all of these functions may be realized by hardware circuits such as ICs.

[0074] Additionally, the control system (85) is equipped with an input unit (85a) composed of a keyboard, mouse, touch panel, etc., a first display unit (85b) and a second display unit (85c) each having a display screen capable of displaying various information composed of a liquid crystal display, etc., a memory unit (85d) capable of storing various data, programs, calculation results, inspection results, etc., and a communication unit (85f) capable of transmitting and receiving various data to and from the outside. In addition, a re-evaluation operation unit (85e) is installed at a location operable by an inspector. In this embodiment, the first display unit (85b) and the second display unit (85c) each constitute a "display means."

[0075] Next, the various functional parts constituting the control system (85) are described in detail.

[0076] The main control unit (851) is a functional unit responsible for controlling the entire judgment unit (8) and is configured to transmit and receive various signals to and from other functional units, such as the lighting control unit (852) and the camera control unit (853).

[0077] The lighting control unit (852) is a functional unit that controls the lighting devices (81, 82) based on a command signal from the main control unit (851).

[0078] The camera control unit (853) is a functional unit that controls the cameras (83, 84) and controls the timing of imaging by the cameras (83, 84) based on a command signal from the main control unit (851). In this embodiment, the camera control unit (853) controls each camera (83, 84) to image the same tablet (9) at least twice in the upstream transport unit (41) and to image the same tablet (9) at least twice in the downstream transport unit (43).

[0079] Additionally, when a predetermined operation is performed on the re-evaluation control unit (85e), the camera control unit (853) controls each camera (83, 84) to capture a range corresponding to the defective tablet (9x) (i.e., a range including the part where the defective tablet (9x) is located) while the tablet (9) is temporarily paused. That is, the camera control unit (853) controls the cameras (83, 84) to obtain re-evaluation image data. In this embodiment, the camera control unit (853) comprises an "imaging control means."

[0080] The display control unit (854) controls the display content in each display unit (85b, 85c) based on information stored in the memory unit (85d). In this embodiment, the display control unit (854) constitutes a "display control means."

[0081] The image acquisition unit (855) is a functional unit for receiving image data captured and acquired by each camera (83, 84).

[0082] The data processing unit (856) is a functional unit that performs a predetermined image processing on image data acquired by the image acquisition unit (855). In this embodiment, the data processing unit (856) performs processing to acquire original image data (XD) or learning data (SD) (see FIG. 7 and 8) by extracting areas corresponding to a plurality of refinements (9) in one column from the global data (AD) (image data representing the entire field of view of the camera (83, 84) for each column of refinements (9). The original image data (XD) is used for determining the quality of the refinements (9), and the learning data (SD) is used for learning the neural network (90) described later. The original image data (XD) and learning data (SD) can be acquired in multiple numbers (in this embodiment, equal to the number of columns of refinements (9)) from the global data (AD).

[0083] The learning unit (857) is a functional unit that uses learning data (SD) to perform learning of a deep neural network (90) (hereinafter referred to simply as "neural network (90); see FIG. 9") and builds an AI (Artificial Intelligence) model (100) as an "identification means."

[0084] In addition, the AI ​​model (100) in this embodiment is a generative model constructed by deep learning a neural network (90) using only image data corresponding to the inspection area (KA) of the refined product (9) as training data (SD), as described below, and has the structure of a so-called autoencoder.

[0085] Here, the structure of the neural network (90) is described with reference to FIG. 9. FIG. 9 is a schematic diagram conceptually showing the structure of the neural network (90). As shown in FIG. 9, the neural network (90) has the structure of a convolutional auto-encoder (CAE) consisting of an encoder unit (91) as an "encoding unit" that extracts features (latent variables) (TA) from input image data (GA), and a decoder unit (92) as a "decoding unit" that reconstructs image data (GB) from the features (TA).

[0086] Since the structure of the convolution autoencoder is known, a detailed description is omitted, but the encoder unit (91) has a plurality of convolution layers (93), and in each convolution layer (93), the result of a convolution operation using a plurality of filters (kernels) (94) on the input data is output as input data for the next layer. Likewise, the decoder unit (92) has a plurality of deconvolution layers (95), and in each deconvolution layer (95), the result of a deconvolution operation using a plurality of filters (kernels (96)) on the input data is output as input data for the next layer. Then, in the learning process described later, the weights (parameters) of each filter (94, 96) are updated.

[0087] Returning to FIG. 4, the inspection unit (858) is a functional unit that performs a quality determination of the tablet (9) being transported by the transport device (4). For example, in this embodiment, an inspection is performed to determine whether foreign matter or contamination is attached to the tablet (9), and whether damage such as defects or cracks has occurred in the tablet (9). In addition, if a printing unit is installed on the tablet (9), the inspection unit (858) can perform a quality determination of the printing unit. In this embodiment, since the same tablet (9) is captured at least twice by each camera (83, 84) as described above, the quality determination of the surface (9a) of the tablet (9) is performed at least twice, and the quality determination of the back surface (9b) of the tablet (9) is performed at least twice. In this embodiment, the inspection unit (858) constitutes a "determination means."

[0088] Additionally, the surface (9a) and back surface (9b) of the same tablet (9) are captured at least twice, primarily to increase the precision in determining the quality of the tablet (9). However, if the tablet (9) is defective, the first capture may be performed to detect the defective tablet (9x), and the second capture may be performed to obtain image data of the defective tablet (9x) while it is temporarily stopped. The image data of the defective tablet (9x) while it is temporarily stopped is used to identify the defective tablet (9x) in the judgment result corresponding processing of the step (S305) described later. By using such image data, it becomes possible to almost match the state of the tablet (9) (position, spacing, direction, etc.) indicated by the image data with the state of the actual tablet (9) while it is temporarily stopped.

[0089] Additionally, when a predetermined operation is performed on the re-evaluation control unit (85e), the inspection unit (858) re-evaluates the quality of the tablet (9) based on the re-evaluation image data described above.

[0090] The first display unit (85b) and the second display unit (85c) are positioned so that they are visible to an inspector performing a visual inspection, such as near a conveying device (4), for example (see FIG. 2). In this embodiment, the first display unit (85b) is installed in correspondence with the upstream conveying unit (41), and the second display unit (85c) is installed in correspondence with the downstream conveying unit (43). By controlling each display unit (85b, 85c) with a display control unit (854), the first display unit (85b) displays a result of a good / bad judgment based on image data obtained by the first camera (83), and the second display unit (85c) displays a result of a good / bad judgment based on image data obtained by the second camera (84). In addition, the placement location of each display unit (85b, 85c) can be appropriately changed as long as the inspector can check the information displayed on the display unit (85b, 85c). Of course, each display unit (85b, 85c) may be fixed or portable. In addition, the number of display units installed can also be appropriately changed according to, for example, the number of inspectors.

[0091] The memory unit (85d) is composed of an HDD (Hard Disk Drive) or an SSD (Solid State Drive), and has a predetermined memory area for storing, for example, an AI model (100) (a neural network (90) and learning information obtained through the learning thereof). In addition, the memory unit (85d) is equipped with a function to store image data obtained by each camera (83, 84) or the results of a quality judgment of the refinement (9) performed by the inspection unit (858).

[0092] The re-evaluation operation unit (85e) is an operation device used to re-execute the determination of the quality of the tablet (9) by the inspection unit (858) when the return of the tablet (9) is temporarily suspended. When a predetermined operation is performed on the re-evaluation operation unit (85e), as described above, image data for re-evaluation is obtained by the camera (83, 84), and at the same time, the determination of the quality of the tablet (9) based on the image data for re-evaluation is performed by the inspection unit (858).

[0093] The communication unit (85f) is configured to transmit and receive various data to and from the outside, for example, by being equipped with a wireless communication interface conforming to communication standards such as a wired LAN (Local Area Network) or a wireless LAN. For example, the inspection results performed by the inspection unit (858) are output to the outside (transport control device (7), etc.) through the communication unit (85f).

[0094] Next, the learning process of the neural network (90) performed by the decision unit (8) will be explained with reference to the flowchart of FIG. 10.

[0095] When learning processing is initiated based on the execution of a predetermined learning program, the main control unit (851) performs preprocessing for learning the neural network (90) in the first step (S101).

[0096] In this preprocessing, first, a plurality of quality judgment information stored in the memory unit (85d) is acquired. Subsequently, based on the quality judgment information, image data corresponding to the refined product (9) determined to be good in the quality judgment is acquired from the memory unit (85d). This image data is global data (AD) representing the entire inspection area (KA). Additionally, the data processing unit (856) generates training data (SD) by extracting areas corresponding to a plurality of refined products (9) in one column from the image data (global data (AD)) for each column of refined products (9). The training data (SD) has the same format as the original image data (XD) and is used to perform training of the neural network (90) as described above. Such processing is performed until the required number of training data (SD) is acquired.

[0097] In addition, as training data (SD), virtual good product image data generated using image data according to the refinement (9) of a good product can be used.

[0098] When the required number of training data (SD) is obtained in step (S101), in the next step (S102), the learning unit (857) prepares an unlearned neural network (90) based on a command from the main control unit (851). For example, a neural network (90) that has been previously stored in a memory unit (85d), etc. is read. Alternatively, a neural network (90) is constructed based on network configuration information (for example, the number of layers of the neural network or the number of nodes in each layer) stored in the memory unit (85d), etc.

[0099] In step (S103), reconstructed image data is acquired. That is, based on a command from the main control unit (851), the learning unit (857) applies the learning data (SD) acquired in step (S101) as input data to the input layer of the neural network (90), thereby acquiring the reconstructed image data output from the output layer of the neural network (90). The reconstructed image data has the same format as the learning data (SD) or the original image data (XD) and is image data representing an area corresponding to one column of refinement (9).

[0100] In the next step (S104), the learning unit (857) compares the input image data (learning data (SD)) with the reconstructed image data output by the neural network (90) and determines whether the error is sufficiently small (whether it is below a predetermined threshold).

[0101] Here, if the error is sufficiently small, in step (S106), the learning unit (857) determines whether the learning termination condition is satisfied. For example, if a positive determination in step (S104) is performed consecutively a predetermined number of times without undergoing the processing of step (S105) described later, or if learning using all of the prepared learning data (SD) is performed repeatedly a predetermined number of times, it is determined that the termination condition is satisfied. If the termination condition is satisfied, the neural network (90) and its learning information (parameters after update described later, etc.) are stored in the memory unit (85d) as an AI model (100), and the learning process is terminated.

[0102] Meanwhile, if the termination condition is not satisfied in step (S106), return to step (S103) and re-learn the neural network (90).

[0103] Additionally, if the error is not sufficiently small in step (S104), the network update processing (learning of the neural network (90)) is performed in step (S105), and then the process returns to step (S103) and the series of processes is repeated.

[0104] Specifically, in the network update processing of step (S105), a known learning algorithm such as backpropagation is used, for example, to update the weights (parameters) of each filter (94, 96) in the neural network (90) to more appropriate ones so that the loss function representing the difference between the training data (SD) and the reconstructed image data becomes as small as possible. Additionally, as the loss function, for example, Binary Cross-entropy (BCE) can be used.

[0105] By repeating the processing of steps (S103~S105) several times, the neural network (90) minimizes the error between the training data (SD) and the reconstructed image data, thereby producing more accurate reconstructed image data.

[0106] And, the AI ​​model (100) obtained finally generates reconstructed image data that is almost identical to the input original image data (XD) when image data (in this embodiment, original image data (XD)) according to the purification (9) of a good product is input. In addition, when image data (original image data (XD)) according to the purification (9x) of a defective product is input, the AI ​​model (100) generates reconstructed image data that is almost identical to the original image data (XD) with the noise portion (a portion corresponding to the defective portion) removed. That is, when the purification (9) is defective, virtual image data according to the purification (9) is generated as reconstructed image data according to the purification (9) in the case where it is assumed that there is no defective portion.

[0107] Next, the processing based on the quality judgment of the tablet (9) performed by the judgment unit (8) will be explained with reference to the flowchart of FIG. 11. This processing is performed at a preset normal quality judgment execution timing or when a predetermined operation is performed on the re-evaluation operation unit (85e).

[0108] First, in step (S301), imaging of the refinement (9) is performed by each camera (83, 84). That is, by controlling each camera (83, 84) through the camera control unit (853), the inspection area (KA) of each return unit (41, 43) is captured by each camera (83, 84). By doing so, image data (global data (AD)) representing the entire inspection area (KA) is obtained.

[0109] In the next step (S302), the data processing unit (856) performs a process of extracting (cutting out) areas corresponding to one column of multiple refinements (9) from the obtained global data (AD) for each column of refinements (9). In step (S302), one column of original image data (XD) is obtained from the global data (AD).

[0110] In step (S303), a pass / fail determination process is performed. Specifically, as illustrated in FIG. 12, first, in step (S501), the original image data (XD) acquired in step (S302) by the inspection unit (858) is input to the input layer of the AI ​​model (100). As a result, reconstructed image data is output from the AI ​​model (100). In this embodiment, the inspection unit (858) constitutes a "reconstructed image data acquisition means."

[0111] In step (502), the original image data (XD) and the reconstructed image data are compared by the inspection unit (858), and the difference between the two image data is calculated. For example, dots at the same coordinates in the two image data are compared, and the area (number of dots) of the dot clusters where the difference in brightness is greater than or equal to a predetermined value is calculated. In this embodiment, the inspection unit (858) that compares the original image data (XD) and the reconstructed image data constitutes the "comparison means."

[0112] Next, in step (S503), the inspection unit (858) determines whether there are any defective parts in the tablet (9). Specifically, the inspection unit (858) determines whether each calculated difference is greater than a predetermined threshold. Then, if the difference is greater than the predetermined threshold, it is determined to be "defective" in step (S504). On the other hand, if the difference is less than the predetermined threshold, it is determined to be "good" in step (S505).

[0113] Returning to Fig. 11, in step (S304) which continues the process of determining the quality of step (S303), it is determined whether the process of determining the quality of step (S303) has been performed based on all original image data (XD) according to the image data (global data (AD)) obtained in step (S301). That is, in step (S301), a plurality of original image data (XD) (equal to the number of columns of the refinement (9)) are obtained from the global data (AD), and it is determined whether the process of determining the quality of step (S303) has been performed based on all of these original image data (XD).

[0114] Here, if a pass / fail judgment process based on all original image data (XD) is performed (step (S304): YES), proceed to step (S305). On the other hand, if not (step S304: NO), return to step (S302). Therefore, the processing of steps (S302, S303) is repeated until a pass / fail judgment process based on all original image data (XD) is performed.

[0115] Then, when a quality judgment based on all original image data (XD) is performed, in the processing corresponding to the judgment result of step (S305), processing according to the result of the quality judgment is executed. That is, as a result of the quality judgment processing performed based on all original image data (XD), if all original image data (XD) is judged to be "good," the tablet (9) according to the inspection area (KA) of the image target is judged to be "good," and this judgment result is stored in the memory unit (85d). In addition, information indicating that it is "good" and the image data (global data (AD)) obtained in step (S301) are displayed on the display unit (85b, 85c) by the display control unit (854).

[0116] Meanwhile, if at least one original image data (XD) is determined to be "defective," the tablet (9) according to the inspection area (KA) of the image target is determined to be "defective," and the determination result and information indicating the location of the defective part (e.g., coordinate information, etc.) are stored in the memory unit (85d).

[0117] Additionally, information indicating "defective product" and information for specifying the defective tablet (9x) is displayed on the display unit (85b, 85c) in correspondence with the image data (global data (AD)) obtained in step (S301) by the display control unit (854). In this embodiment, based on the information indicating the location of the defective part described above, a mark (MK) indicating the location is displayed in a state overlapping the image data (global data (AD)) obtained in step (S301) (see FIG. 13). That is, a specific mark (MK) indicating the location of the defective tablet (9x) and the location of the defective part in the defective tablet (9x), respectively, is displayed in a state overlapping the image data (global data (AD)).

[0118] In addition, when the tablet (9) is determined to be a “defective product” based on the image data obtained from the first imaging of any tablet (9), the transport of the tablet (9) is not temporarily stopped immediately after this determination, and when the timing is reached to perform the second imaging of the defective tablet (9x), the transport of the tablet (9) is temporarily stopped by the transport control device (7).

[0119] Then, while the return is temporarily paused, a second image of the defective tablet (9x) is taken, and simultaneously, a pass / fail judgment is made based on the image data obtained from this image. In this pass / fail judgment, it is usually judged to be a "defective product," just like the first pass / fail judgment. Then, in the display unit (85b, 85c), information (mark (MK)) for identifying the defective tablet (9x) is displayed in correspondence with the image data (global data (AD)) obtained from the second image. At this time, since the image data (global data (AD)) displayed in the display unit (85b, 85c) was acquired while the tablet (9) was temporarily paused, the current state of the tablet (9) (position, spacing, direction, etc.) and the state of the tablet (9) indicated by the image data displayed in the display unit (85b, 85c) almost match.

[0120] When the above information is displayed, the defective tablet (9x) is identified by an inspector performing a visual inspection based on this information, and at the same time, the defective tablet (9x) is removed from the return device (4) by means of tweezers, etc. At this time, since the current state of the tablet (9) and the state of the tablet (9) indicated by the image data displayed on the display unit (85b, 85c) are almost identical, the inspector can very easily identify the defective tablet (9x) by using the display content on the display unit (85b, 85c).

[0121] Additionally, after the defective tablet (9x) is removed, the return of the tablet (9) is resumed by the inspector. However, as described above, the return of the tablet (9) is not resumed simply by operating the return operation unit. In order to resume the return of the tablet (9), it is necessary to operate the re-evaluation operation unit (85e) to re-perform the quality judgment process, and then operate the return operation unit after satisfying the condition that the tablet is judged to be a "good product" by this quality judgment process. In addition, the judgment results and image data regarding these two quality judgment processes are also displayed on the display unit (85b, 85c).

[0122] As described above, according to the present embodiment, the quality of the tablet (9) being returned by the return device (4) (i.e., subject to visual inspection) is determined by the inspection unit (858). And, if a defect is determined by the inspection unit (858), the return of the tablet (9) is temporarily stopped. In addition, information for identifying the defective tablet (9x) is displayed in each display unit (85b, 85c).

[0123] Therefore, an inspector performing a visual inspection can effectively utilize the display content in the display section (85b, 85c) to easily identify a defective tablet (9x) among a plurality of tablets (9). In addition, since the tablet (9) is temporarily stopped when a defect is determined, the inspector can remove the specific defective tablet (9x) more accurately. This makes it more reliable to prevent overlooking a defective tablet (9x), mishandling a defective tablet (9x), or mistakenly removing a good tablet (9) that is different from the defective tablet (9x). As a result, it makes it more reliable to prevent the defective tablet (9x) from flowing downstream, thereby increasing the reliability of the visual inspection.

[0124] Additionally, in the display unit (85b, 85c), information for identifying a defective tablet (9x) (mark (MK) in this embodiment) is displayed in correspondence with image data corresponding to the defective tablet (9x). Accordingly, identifying the defective tablet (9x) becomes easier, and the effort or time required to remove the defective tablet (9x) can be further reduced.

[0125] In addition, when a defective tablet (9x) is discovered and the return of the tablet (9) is temporarily suspended, the return device (4) can be restarted and the return of the tablet (9) resumed only if the judgment result based on the re-evaluation image data obtained by capturing at least the area corresponding to the defective tablet (9x) (i.e., the area including the part where the defective tablet (9x) is located) is satisfactory. Therefore, even if the defective tablet (9x) is overlooked or a mistake occurs, it is more reliably prevented from being accidentally discharged downstream. As a result, the reliability of the visual inspection can be further improved.

[0126] In addition, as training data (SD) or original image data (XD), image data obtained by extracting areas corresponding to multiple refinements (9) in one column adjacent to each other in the return direction for each column of refinement (9) is used. Accordingly, the amount of data required for training can be reduced relatively, and the cost associated with training can be reduced. In addition, the extraction processing of training data (SD) or original image data (XD) can be made easier.

[0127] In addition, the original image data (XD) and the reconstructed image data, which is reconstructed by inputting the original image data (XD) into the AI ​​model (100), are compared, and the quality of the tablet (9) is determined based on the comparison result. Therefore, the two image data being compared each relate to the same tablet (9). Consequently, since the shape and appearance of the tablet (9) in the two image data being compared are almost identical, there is no need to set relatively loose judgment conditions to prevent false detection due to differences in shape or appearance, and stricter judgment conditions can be set. Furthermore, the imaging conditions (e.g., the placement position or placement angle of the tablet (9), brightness and contrast state, or camera angle of view, etc.) in the two image data being compared can be matched. As a result, the quality of the tablet (9) can be determined with great precision.

[0128] In addition, in this embodiment, since an inversion part (42) that reverses the front and back sides of the tablet (9) is installed, visual inspection of the front side (9a) and back side (9b) of the tablet (9) can be easily performed.

[0129] Furthermore, the above embodiments are not limited to the descriptions provided and may be implemented as follows, for example. Of course, other applications and modifications not exemplified below are also naturally possible.

[0130] (a) In the above embodiment, each camera (83, 84) is positioned above the transport path of the tablet (9) and is configured to capture the upper surface of the tablet (9) respectively.

[0131] In this regard, as illustrated in FIGS. 14 and 15, a transparent or translucent transparent portion (46) (the portion shown as a scattered dot in FIG. 15) is installed on at least a part of the tablet support portion (45), and at the same time, each camera (83, 84) is positioned to fit the tablet support portion (45) up and down, and can be configured to capture the upper surface of the tablet (9) by the first camera (83) and the lower surface of the tablet (9) through the transparent portion (46) by the second camera (84). That is, it can be configured to capture one of the surface (9a) and the back surface (9b) of the tablet (9) by the first camera (83), and to capture the other of the surface (9a) and the back surface (9b) of the tablet (9) through the transparent portion (46) by the second camera (84). And, the inspection unit (858) can be configured to determine the quality of the tablet (9) based on each image data according to the surface (9a) and back side (9b) of the tablet (9) obtained by each camera (83, 84).

[0132] In the case configured as above, the second camera (84) is positioned below the tablet support (45), so that the second camera (84) can be prevented from interfering with visual inspection.

[0133] In addition, the return device (4) is allowed to start (restart) only when both the surface (9a) and the back side (9b) of the tablet (9) are determined to be good based on image data for re-evaluation. Because of this, even if a defective tablet (9x) is mishandled and the defective tablet (9x) is turned over, the return device (4) is not allowed to start (restart). Therefore, the defective tablet (9x) can be very effectively prevented from flowing downstream.

[0134] (b) In the above embodiment, as original image data (XD) and training data (SD), an image data (global data (AD)) representing the entire area to be examined (KA) is used, and an area corresponding to a plurality of refinements (9) in one column is extracted for each column of refinements (9). However, the global data (AD) or image data obtained by extracting one area each of refinements (9) from the global data (AD) can be used.

[0135] (c) In the above embodiment, the inspection unit (858) is configured to determine the quality of the surface (9a) and back side (9b) of the tablet (9). In addition, the inspection unit (858) may be configured to determine the quality of the side side (9c) of the tablet (9).

[0136] For example, when the front and back sides of the tablet (9) are inverted by the inversion unit (42), the side (9c) of the tablet (9) is in a state facing upward, and a side camera (86) is installed to capture the side (9c) of the tablet (9) located in the inversion unit (42) as shown in FIG. 16, and the inspection unit (858) can be configured to determine the quality of the side (9c) based on the image data obtained by the side camera (86). The side camera (86) corresponds to a "side imaging means" and an "imaging means."

[0137] When configured in this way, the inversion unit (42) can be used to determine the quality of the side (9c) of the tablet (9). Accordingly, the precision of the quality determination of the tablet (9) can be further increased. In addition, it can further prevent defective tablets (9x) from flowing downstream and increase the reliability of the visual inspection.

[0138] (d) A cover may be installed to cover all or part of the upstream return section (41), the inversion section (42), and the downstream return section (43). Additionally, a camera (imaging means) may be installed within the cover, and the configuration may be such that a quality determination of the tablet (9) is performed based on image data obtained by the camera.

[0139] In addition, the state of the cover can be used as a condition for resuming the return of the tablet (9) or as a condition for taking an image by a camera (83, 84) to acquire image data for re-review. For example, the condition that the cover is properly installed or that the opening installed in the cover is closed can be used as a condition for resuming the return or taking an image.

[0140] (e) In the above embodiment, the camera (83, 84) is configured to capture the tablet (9) of 1 at least twice each, but the number of times the tablet (9) of 1 is captured can be appropriately changed.

[0141] (f) In the above embodiment, when a defect judgment of the tablet (9) is made based on the image data obtained from the first imaging, the transport of the tablet (9) is temporarily stopped at the timing of the second imaging of the tablet (9), but the timing of the temporary stop can be appropriately changed. For example, the transport of the tablet (9) can be temporarily stopped immediately after the defect judgment.

[0142] In addition, when configured in this way, since a little time is required from the execution of the image to the completion of the pass / fail judgment, when the return of the tablet (9) is stopped, the actual position of the defective tablet (9x) may differ slightly from the position of the defective tablet (9x) in the image data (global data (AD)) used for the pass / fail judgment.

[0143] Accordingly, for example, the average return amount of the tablet (9) during the period from the execution of the image to the completion of the pass / fail determination can be obtained and estimated in advance, and as shown in FIG. 17, the display unit (85b, 85c) can be configured to display a mark (MK1) indicating a downstream position by the average return amount from the defective tablet (9x) in correspondence with the image data (global data (AD)) corresponding to the defective tablet (9x). Even when configured in this way, the identification of the defective tablet (9x) can be performed relatively easily. In addition, considering that an error occurs between the actual return amount and the average return amount, a mark indicating a relatively wide area following the return direction of the tablet (9) can be used as the mark (MK1).

[0144] In addition, for example, as shown in FIG. 18, a mark (MK2) indicating a column containing defective purification (9x) can be displayed in correspondence with image data (global data (AD)) corresponding to the defective purification (9x). Even in such a configuration, the identification of defective purification (9x) can be performed relatively easily.

[0145] In addition, it is not necessary to display information for identifying the defective tablet (9x) in correspondence with image data corresponding to the defective tablet (9x). For example, as shown in FIG. 19, the configuration may be such that a mark (MK3) indicating the location of the defective tablet (9x) is displayed within a frame (WK) corresponding to the inspection area (KA) (the imaging range of the camera (83, 84, 86)). In addition, in this case, by drawing an outline indicating the inspection area (KA) (the imaging range of the camera (83, 84, 86)) on the upper surface of the transport device (4) (such as the upstream transport unit (41) or the downstream transport unit (43)), the identification of the defective tablet (9x) based on the displayed content can be performed even more easily. In addition, an auxiliary line (ML), etc., corresponding to the transport groove (41a, 42a, 43a) may be installed within the frame (WK).

[0146] (g) In the above embodiment, a flat tablet (9) in the shape of a disc that forms a circular shape when viewed from a flat surface is exemplified, but the type and shape of the tablet are not limited to the above embodiment. For example, the tablet includes not only medicine but also tablets used for food. In addition, the tablet includes not only the prescribed tablet but also coated tablets, film-coated tablets, oral disintegrating tablets, enteric-coated tablets, gelatin-coated tablets, etc., as well as various capsule tablets such as hard capsules and soft capsules.

[0147] In addition, regarding the shape of the tablet (9), for example, when viewed from a flat plane, it may be not only circular, but also polygonal, elliptical, or circular.

[0148] (h) In the above embodiment, the tablet (9) is configured to be supplied from the visual inspection aid device (1) (chute (5)) to the counter (6a) and the bottle filling device (6b), but the supply target of the tablet (9) can be appropriately changed. Accordingly, for example, a blister packaging machine (e.g., PTP packaging machine) for manufacturing a blister sheet (e.g., PTP sheet) in which the tablet (9) is received in a pocket formed in a container film and a cover film is attached to the container film to block the pocket can be used as the supply target of the tablet (9). Additionally, the blister packaging machine can be configured to have the visual inspection aid device (1).

[0149] (i) The configuration of the AI ​​model (100) (neural network (90)) as an "identification means" and the method of learning the same are not limited to the above embodiments. For example, when performing the learning processing of the neural network (90) or the acquisition processing of reconstructed image data, the configuration may be configured to perform processing such as normalization on various data as needed. In addition, the structure of the neural network (90) is not limited to that shown in FIG. 9, and may be configured, for example, by installing a pooling layer after the convolution layer (93). Of course, the number of layers of the neural network (90), the number of nodes in each layer, and the connection structure of each node may be configured differently.

[0150] In addition, in the above embodiment, the AI ​​model (100) (neural network (90)) is a generative model having the structure of a convolutional autoencoder (CAE), but is not limited thereto and can be a generative model having the structure of another type of autoencoder, such as a variational autoencoder (VAE).

[0151] In addition, in the above embodiment, the neural network (90) is configured to be trained by the error backpropagation method, but it is not limited thereto and can be configured to be trained using various other learning algorithms.

[0152] Additionally, the neural network (90) may be configured by a circuit dedicated to AI processing, such as a so-called AI chip. In this case, only learning information such as parameters is stored in the memory unit (85d), and the circuit dedicated to AI processing reads this and sets it in the neural network (90), thereby enabling the configuration of an AI model (100).

[0153] In addition, in the above embodiment, the control system (85) is configured to have a learning unit (857) and to perform learning of the neural network (90) within the control system (85), but is not limited thereto. For example, the learning unit (857) may be omitted, and the learning of the neural network (90) may be performed outside the control system (85), and the AI ​​model (100) (the neural network (90) that has completed learning) that has been trained outside may be stored in the memory unit (85d). Of course, the learning data (SD) may be generated outside the control system (85).

[0154] (j) The configuration of the conveying device (4) is not limited to the example given in the above embodiment, and the conveying device can be configured by, for example, a conveyor, etc. Explanation of the symbols

[0155] 1: Visual inspection aid 4: Return device (return means) 7: Return control device (return control means) 9: Refining 9a: Surface 9b: If 9c… side 42: Counter-incident 45: Tablet support 46: Transmissive part 83: First camera (imaging means, first imaging means) 84: Second camera (imaging means, second imaging means) 85b: First display part (display means), 85c: Second display part (second display means), 86: Side camera (imaging means, side imaging means) 90: Neural Network 91: Encoder section (Encoding section) 92: Decoder (Decoding) 100: AI model (means of identification) 853: Camera judgment unit (imaging control means) 854: Display control unit (display control means) 858: Inspection unit (determination means, means for acquiring reconstructed image data, comparison means)

Claims

Claim 1 A visual inspection auxiliary device comprising a conveying means for conveying a tablet subject to visual inspection, the device comprises: an imaging means for capturing an image of a tablet conveyed by the conveying means; a judgment means for determining the quality of a tablet based on image data obtained by the imaging means; a display means positioned at a location visible to an inspector performing a visual inspection of the tablet; a display control means capable of displaying information for identifying a tablet that has been judged defective by at least the judgment means in the display means; and a conveying control means for temporarily stopping the conveying of the tablet by the conveying means when a defect is judged defective by the judgment means, wherein the device is configured to identify and remove a tablet that has been judged defective by using the display content in the display means, wherein the tablet has a surface and a back surface, and the conveying means has a tablet support member that supports the conveyed tablet while having a transparent or translucent transparent portion installed in at least a part thereof, and the imaging means comprises a first imaging means and a second imaging means positioned at a location that inserts the tablet support member vertically. A visual inspection assist device characterized by having, wherein one side of the surface and the back side of a tablet is imaged by the first imaging means and the other side of the surface and the back side of the tablet is imaged through the transmission part by the second imaging means, and the determination means is configured to determine the quality of the tablet based on image data according to the surface and back side of the tablet obtained by the first imaging means and the second imaging means. Claim 2 A visual inspection assisting device having a conveying means for conveying a tablet subject to visual inspection, the device comprises: an imaging means for capturing an image of a tablet conveyed by the conveying means; a judgment means for determining the quality of a tablet based on image data obtained by the imaging means; a display means positioned at a location visible to an inspector performing a visual inspection of the tablet; a display control means capable of displaying information for identifying a tablet that has been judged defective by at least the judgment means in the display means; and a conveying control means for temporarily stopping the conveying of the tablet by the conveying means when a defect is judged defective by the judgment means; and an imaging control means for controlling the imaging means, wherein the imaging control means controls the imaging means to capture at least a range corresponding to the tablet judged defective during the temporary suspension of the tablet to obtain re-evaluation image data, which is the image data for re-evaluation, and simultaneously, the judgment means determines the quality of the tablet based on the re-evaluation image data, and the conveying control means allows the conveying of the tablet that is temporarily suspended to resume. A visual inspection assist device characterized by the ability to restart a return means, wherein the restart of the return means by the return control means is configured to be performed under the condition that the judgment result by the judgment means based on the re-evaluation image data is good. Claim 3 A visual inspection assisting device equipped with a conveying means for conveying tablets subject to visual inspection, the device comprises: an imaging means for capturing a tablet conveyed by the conveying means; a judgment means for determining the quality of a tablet based on image data obtained by the imaging means; a display means positioned at a location visible to an inspector performing a visual inspection of the tablet; a display control means capable of displaying information for identifying a tablet that has been judged defective by at least the judgment means; and a conveying control means for temporarily stopping the conveying of the tablet by the conveying means when a defect is judged defective by the judgment means. The conveying means conveys the tablets in a state arranged in multiple rows. The judgment means comprises: an identification means generated by training only learning data based on image data of a good tablet for a neural network having an encoding unit that extracts feature quantities from input image data and a decoding unit that reconstructs image data from the feature quantities; and an image reconstructed by inputting original image data based on image data obtained by the imaging means into the identification means. A visual inspection assisting device comprising: a means for acquiring reconstructed image data capable of acquiring data as reconstructed image data; and a comparison means capable of comparing the original image data and the reconstructed image data, configured to determine the quality of a tablet based on a comparison result by the comparison means; wherein the learning data is image data obtained by extracting an area corresponding to a plurality of tablets in one column from image data of a good product tablet for each column of tablets, and the original image data is image data obtained by extracting an area corresponding to a plurality of tablets in one column from image data obtained by the imaging means for each column of tablets. Claim 4 A visual inspection assist device according to claim 2 or 3, wherein the tablet has a surface and a back surface, the conveying means has a tablet support member that supports the conveyed tablet and is installed with a transparent or translucent transparent portion in at least a part thereof, the imaging means has a first imaging means and a second imaging means positioned to fit the tablet support member up and down, and is configured such that one of the surface and back surface of the tablet is imaged by the first imaging means and the other of the surface and back surface of the tablet is imaged through the transparent portion by the second imaging means, and the judgment means is configured to determine the quality of the tablet based on image data according to the surface and back surface of the tablet obtained by the first imaging means and the second imaging means. Claim 5 A visual inspection assisting device according to one of claims 1 to 3, wherein the display control means is configured to be capable of displaying information for identifying a tablet that has been judged defective by at least the judgment means in the display means, corresponding to the image data of the tablet. Claim 6 A visual inspection assisting device according to any one of claims 1 to 3, wherein the tablet has a surface and a back surface and a side surface located between the surface and the back surface, the conveying means has an inversion part that inverts the front and back surfaces of the conveyed tablet, the imaging means has a side imaging means that images the side surface of the tablet located in the inversion part, and the judgment means is configured to determine the quality of the side surface of the tablet based on image data obtained by the side imaging means.

Images